Learning method, classification method, learning apparatus, classification method, learning program, and classification program

Abstract

To provide a learning apparatus, and the like, capable of training a model while suppressing variability in performance of classifying ambiguous data.SOLUTION: A learning apparatus for classifying input data, using a model, into one of a plurality of classes includes: an acquisition unit for acquiring a reference value of data belonging to a specific class of the multiple classes; a calculation unit which calculates a difference between the reference value and training data; and a training unit which trains the model, while putting more emphasis on the training data with higher degree of importance, the degree of importance being lower when the calculated difference is smaller than when the difference is larger.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The disclosed technology relates to a learning method, a classification method, a learning device, a classification method, a learning program, and a classification program. [Background technology]

[0002] There is a known method for performing segmentation using deep learning (Non-Patent Document 1), in which loss is calculated by ignoring regions that are deemed ambiguous or difficult.

[0003] In addition, a method is known for calculating deep learning loss by weighting each classification class (Non-Patent Document 2).

[0004] There is also known a cell image analysis method for obtaining abnormal cell regions, which are regions in a cell image where the index value indicating that the cell is an abnormal cell rather than a normal cell is greater than a predetermined reference value (Patent Document 1). [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Long, et al., "Fully Convolutional Networks for Semantic Segmentation.", CVPR, 2015. [Non-Patent Document 2] Xie, et al., "Holistically-nested edge detection.", ICCV, 2015. [Patent documents]

[0006] [Patent Document 1] WO2023 / 008526 Summary of the Invention [Problem to be solved by the invention]

[0007] Here, we consider the task of classifying whether input data is an abnormal signal or not. For example, in medical images, segmenting areas with higher or lower signals than normal tissue is useful for quantifying lesions and assisting diagnosis. In addition, in time-series data such as voice and sensor data, extracting the range where there are abnormal signal value changes is useful for anomaly detection in factories, but it is necessary not to judge changes due to noise as abnormal. In general, for ambiguous signals with small differences from normal signal values, the variation in classification performance can be stabilized by lowering the importance and learning, but the definition of ambiguous areas is likely to vary depending on the person.

[0008] In one aspect, an object of the present invention is to provide a learning method, a learning device, and a learning program capable of learning a model while suppressing the variation in classification performance for ambiguous data.

[0009] Another object of the present invention is to provide a classification method, a classification device, and a classification program that can suppress variations in classification performance for ambiguous data. [Means for solving the problem]

[0010] A first aspect of the present disclosure is a learning method for classifying input data into which of a plurality of classes the data belongs using a model, wherein a computer executes processes including the steps of obtaining a reference value for data belonging to a specific class among the plurality of classes, calculating the difference between the reference value and the learning data, and, when the calculated difference is small, assigning a lower importance to the learning data than when the difference is large, and training the model by placing more importance on the learning data with a higher importance.

[0011] A second aspect of the present disclosure is a classification method including a step of classifying input data into which of the multiple classes it belongs using a trained model trained by the learning method of the first aspect.

[0012] A third aspect of the present disclosure is a learning device for classifying input data into which of a plurality of classes the data belongs using a model, the learning device including: an acquisition unit that acquires a reference value of data belonging to a specific class among the plurality of classes; a calculation unit that calculates the difference between the reference value and learning data; and a learning unit that, when the calculated difference is small, assigns a lower importance to the learning data than when the difference is large, and trains the model by placing more importance on the learning data with a higher importance.

[0013] A fourth aspect of the present disclosure is a classification device that includes a classification unit that classifies input data into which of the multiple classes the data belongs, using a trained model trained by the learning device of the third aspect.

[0014] A fifth aspect of the present disclosure is a learning program for classifying input data into which of a plurality of classes the data belongs using a model, the program causing a computer to execute a process of learning the model by obtaining a reference value for data belonging to a specific class among the plurality of classes, calculating the difference between the reference value and the learning data, and assigning a lower importance to the learning data when the calculated difference is small compared to when the difference is large, and giving more importance to the learning data with a higher importance.

[0015] A sixth aspect of the present disclosure is a classification program that causes a computer to execute a process of classifying input data into which of the multiple classes the data belongs, using a trained model trained by the learning program of the fifth aspect. Effect of the Invention

[0016] In one aspect, a model can be trained with reduced variability in classification performance for ambiguous data.

[0017] In addition, the variation in classification performance for ambiguous data can be suppressed. [Brief description of the drawings]

[0018] [Figure 1] FIG. 1 is an explanatory diagram illustrating an example of the configuration of a classification system. [Diagram 2] FIG. 1 is a schematic block diagram of an example of a computer that functions as a learning device and a classification device of the present embodiment. [Diagram 3] 1 is a block diagram showing a configuration of a learning device according to an embodiment of the present invention; [Figure 4] FIG. 13 is a diagram illustrating an example of setting importance levels. [Diagram 5] FIG. 13 is a diagram illustrating an example of setting importance levels. [Figure 6] FIG. 13 is a diagram illustrating an example of setting importance levels. [Figure 7] FIG. 13 is a diagram illustrating an example of setting importance levels. [Figure 8] FIG. 13 is a diagram illustrating an example of setting importance levels. [Figure 9] 1 is a block diagram showing a configuration of a classification device according to an embodiment of the present invention. [Figure 10] 4 is a flowchart showing a learning process routine in the learning device of the present embodiment. [Figure 11] 4 is a flowchart showing a classification processing routine in the classification device of the present embodiment. [Figure 12] FIG. 1 is a diagram illustrating an example of a medical image. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0019] Hereinafter, an example of an embodiment of the disclosed technology will be described with reference to the drawings. In each drawing, the same or equivalent components and parts are given the same reference numerals. In addition, the dimensional ratios of the drawings are exaggerated for the convenience of explanation and may differ from the actual ratios.

[0020] <System configuration> 1 is an explanatory diagram showing an example of the configuration of a classification system 2. In this embodiment, a classification system 2 that performs segmentation of abnormal regions in medical images will be described. The classification system 2 includes a learning device 10 and a classification device 20. The learning device 10 and the classification device 20 are communicatively connected via a network N.

[0021] In this embodiment, the brain is taken as an example of a body part to be classified, but other body parts may be used.

[0022] The learning device 10 is, for example, a server computer capable of various information processing and information transmission and reception. Note that the device corresponding to the learning device 10 is not limited to a server computer, and may be, for example, a personal computer. In this embodiment, the learning device 10 performs machine learning such as CNN (Convolution Neural Network) using previously prepared learning data, and functions as a learning device that learns a model that is a classifier that classifies each pixel as an abnormal region or a normal region when a captured medical image is input. Data of the trained model trained by the learning device 10 is installed in the classification device 20, and the classification device 20 performs segmentation of an abnormal region from a medical image using the trained model.

[0023] The classification device 20 is a general-purpose computer such as a personal computer. In this embodiment, the classification device 20 uses a trained model to classify each pixel in a medical image as being an abnormal region or a normal region, and performs segmentation of the abnormal region.

[0024] In this embodiment, the learning device 10 generates (learns) the trained model, but the local classification device 20 may generate the trained model. In this embodiment, the classification device 20 classifies the region into either an abnormal region or a normal region based on the trained model, but the learning device 10 on the cloud may classify the region into either an abnormal region or a normal region using the trained model. In other words, the distinction between the two is for convenience, and a single computer may perform a series of processes.

[0025] <Configuration of the learning device according to this embodiment> FIG. 2 is a block diagram showing the hardware configuration of the learning device 10 of this embodiment.

[0026] 2, the learning device 10 has a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.

[0027] The CPU 11 is a central processing unit, and executes various programs and controls each part. That is, the CPU 11 reads a program from the ROM 12 or the storage 14, and executes the program using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various arithmetic processing according to the program stored in the ROM 12 or the storage 14. In this embodiment, a learning program for executing a learning process is stored in the ROM 12 or the storage 14. The learning program may be one program, or may be a group of programs composed of multiple programs or modules.

[0028] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including an operating system and various data.

[0029] The input unit 15 includes a pointing device such as a mouse, and a keyboard, and is used to perform various input operations.

[0030] The input unit 15 receives, as learning data, a plurality of medical images to which abnormal regions have been added in advance.

[0031] The display unit 16 is, for example, a liquid crystal display, and displays various information. The display unit 16 may function as the input unit 15 by adopting a touch panel system.

[0032] The communication interface 17 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, and Wi-Fi (registered trademark).

[0033] Next, a description will be given of the functional configuration of the learning device 10. FIG 3 is a block diagram showing an example of the functional configuration of the learning device 10.

[0034] As shown in FIG. 3, the learning device 10 functionally includes a learning data storage unit 101, an acquisition unit 102, a calculation unit 103, and a learning unit 104.

[0035] A plurality of input learning data are stored in the learning data storage unit 101. Specifically, the learning data storage unit 101 stores a plurality of learning medical images to which an abnormal region made up of pixels belonging to an abnormal class is previously assigned.

[0036] The acquiring unit 102 acquires a reference value of data belonging to a specific class among the plurality of classes. In this embodiment, the plurality of classes includes a normal class and an abnormal class, and the normal class is the specific class.

[0037] Specifically, the acquiring unit 102 acquires the reference value by one of the following first to fourth methods.

[0038] In the first method, a reference value is obtained from data that does not satisfy a condition related to a predetermined feature for data classified as belonging to a class other than a specific class. More specifically, an input of a condition related to the feature for the abnormal class is accepted, pixels that do not satisfy the condition related to the feature for the abnormal class are extracted from the medical image for learning, and a reference value for the normal class is obtained from the pixel value of each extracted pixel.

[0039] In the second method, a reference value input from outside is acquired. More specifically, a reference value of a normal class is input.

[0040] In the third method, a representative value is calculated from multiple data belonging to a specific class that are prepared in advance, and is taken as a reference value. More specifically, the pixel values ​​of each pixel belonging to a normal class are obtained from medical images for learning, and a representative value is calculated from the pixel values ​​of each obtained pixel, and is taken as the reference value for the normal class.

[0041] In the fourth method, the maximum or minimum pixel value obtained from an abnormal area belonging to a pre-assigned abnormal class among input data is obtained as a reference value for data belonging to a normal class. More specifically, the pixel value of each pixel belonging to an abnormal class is obtained from a medical image for learning, and the maximum or minimum pixel value is calculated from the pixel values ​​of each obtained pixel, and obtained as the reference value for the normal class.

[0042] The calculation unit 103 calculates the difference between the reference value acquired by the acquisition unit 102 and the learning data stored in the learning data storage unit 101. Specifically, for each pixel of the learning medical image, the calculation unit 103 calculates the difference between the pixel value of the pixel and the reference value.

[0043] The learning unit 104 performs model learning by assigning a lower importance to the learning data stored in the learning data storage unit 101 when the calculated difference is small compared to when the difference is large, and by placing more importance on the learning data with a higher importance.

[0044] Specifically, first, the importance is set by any one of the following first to third methods.

[0045] In the first method, when the calculated difference in the training data is smaller than a threshold, the importance is set to 0 and the model is trained. More specifically, when the absolute value of the calculated difference in each pixel of the training medical image is smaller than a threshold, the importance of the pixel is set to 0 and the model is trained (see FIG. 4). FIG. 4 shows an example in which the importance of a pixel is set to 0 and the pixel is not used for training when the absolute value of the calculated difference is smaller than the threshold. The importance can be used as a value to multiply the learning coefficient or loss function so that the pixel is not used for training.

[0046] In the second method, when the calculated difference of the learning data is smaller than a threshold, the importance of the data is set lower than that of the data whose calculated difference is larger than the threshold, and model learning is performed. More specifically, when the calculated difference of each pixel of the learning medical image is smaller than the threshold, the importance of the pixel is set lower than that of the pixel whose calculated difference is larger than the threshold, and model learning is performed (see FIG. 5). FIG. 5 shows an example in which, when the absolute value of the calculated difference is smaller than the threshold, the importance of the pixel is set lower than that of the pixel whose absolute value of the calculated difference is larger than the threshold, and the importance of the pixel is used as the value by which the loss function is multiplied. In other words, the value by which the loss function is multiplied corresponds to the importance.

[0047] In the third method, a function that monotonically increases with respect to the magnitude of the difference is used for the learning data, and the output value of the function for the calculated difference is used as the importance to learn the model. More specifically, for each pixel of the learning medical image, the output value of the function that monotonically increases as the magnitude of the absolute value of the calculated difference increases is set as the importance of the pixel, and the model is learned (see Figure 6). Figure 6 shows an example in which the output value of the function that monotonically increases as the magnitude of the absolute value of the calculated difference increases is used as the importance of the pixel, and is multiplied by the loss function.

[0048] Furthermore, the learning unit 104 learns the model so as to optimize the value of the loss function calculated based on the importance of the learning data. At this time, the model is generated so that when the learning data is input to the model, the same classification result as that of a region belonging to a previously assigned abnormal class is obtained.

[0049] Specifically, as shown in Figure 7, a medical image for training is used as input, and an output value is obtained for each pixel as the output of the model. The output value for each pixel is compared with the abnormal area consisting of pixels belonging to the abnormal class assigned to the medical image for training to calculate the difference for each pixel. The difference for each pixel is multiplied by the importance of each pixel to calculate the value of the loss function, and the parameters of the model are updated to optimize the value of the loss function.

[0050] Alternatively, the learning unit 104 learns the model using the gradient of the error backpropagation calculated based on the importance of the learning data. Specifically, as shown in Fig. 8, a learning medical image is input, an output value is obtained for each pixel as the output of the model, the output value for each pixel is compared with an area consisting of pixels belonging to an abnormal class assigned to the learning medical image, the difference for each pixel is calculated as the value of the loss function, a gradient is calculated for each pixel so as to optimize the value of the loss function, and the parameters of the model are updated based on the result of multiplying the gradient for each pixel by the importance for each pixel.

[0051] <Configuration of the Classification Device According to the Present Embodiment> FIG. 2 is a block diagram showing the hardware configuration of the classification device 20 of this embodiment.

[0052] 2, the classification device 20, like the learning device 10, has a CPU 11, a ROM 12, a RAM 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other so as to be able to communicate with each other via a bus 19. Note that the classification device 20 preferably includes a processor with high computational processing power such as a GPU in order to perform image processing.

[0053] A classification program for executing classification processing is stored in the ROM 12 or the storage 14. The classification program may be one program, or may be a group of programs configured with multiple programs or modules.

[0054] The input unit 15 receives medical images to be classified.

[0055] Next, a description will be given of the functional configuration of the classification device 20. FIG.

[0056] Functionally, the classification device 20 includes an input acquisition unit 201, a classification unit 202, and an output unit 203, as shown in FIG.

[0057] The input acquisition unit 201 acquires medical images to be classified.

[0058] The classification unit 202 includes a trained model trained by the learning device 10 for classifying which of a plurality of classes an image belongs to. In this embodiment, the trained model outputs a score indicating the degree to which each pixel of an input image belongs to an abnormal class. The trained model is a neural network. It may also be a model based on other learning algorithms, such as SVM, decision tree, or random forest. A higher score indicates a higher possibility of belonging to an abnormal class.

[0059] Specifically, the classification unit 202 uses the trained model to classify each pixel of the medical image into which of the multiple classes the pixel belongs, thereby segmenting an area made up of pixels belonging to at least one of the multiple classes. More specifically, the classification unit 202 uses the trained model to classify each pixel of the medical image into either a normal class or an abnormal class, thereby segmenting an abnormal area made up of pixels belonging to the abnormal class.

[0060] The output unit 203 outputs the segmentation result by the classification unit 202. For example, for a medical image, the display unit 16 displays an abnormal region made up of pixels belonging to an abnormal class.

[0061] <Function of the learning device according to this embodiment> Next, the operation of the learning device 10 according to this embodiment will be described.

[0062] The CPU 11 reads out a learning program from the ROM 12 or storage 14, expands it in the RAM 13, and executes it to perform the learning process shown in Fig. 10. It is also assumed that a plurality of medical images, to which abnormal regions made up of pixels belonging to an abnormal class have been added in advance, are input as learning data to the learning device 10, and that the input plurality of pieces of learning data are stored in the learning data storage unit 101.

[0063] First, in step S100, the CPU 11 functioning as the acquisition unit 102 acquires a reference value of data belonging to a specific class among a plurality of classes.

[0064] In step S102, the CPU 11 functions as the calculation unit 103 to calculate the difference between the reference value acquired by the acquisition unit 102 and the learning data stored in the learning data storage unit 101.

[0065] In step S104, the CPU 11, as the learning unit 104, sets the importance of each pixel of the training medical image by assigning a lower importance to training data stored in the training data memory unit 101 when the calculated difference is small compared to when the difference is large, and by giving more importance to training data with higher importance.

[0066] In step S106, the CPU 11, functioning as the learning unit 104, learns a model so as to optimize the value of a loss function calculated based on the importance of the learning data.

[0067] In step S108, the CPU 11 judges whether or not to end the iteration, and judges to end the iteration and ends the learning process when the number of iterations reaches the upper limit, for example. On the other hand, when the number of iterations does not reach the upper limit, the iteration is not ended and the process returns to step S106.

[0068] <Function of the classification device according to this embodiment> Next, the operation of the classification device 20 according to this embodiment will be described.

[0069] The CPU 11 of the classification device 20 reads out the classification program from the ROM 12 or the storage 14, loads it into the RAM 13, and executes it, thereby performing the classification process shown in FIG.

[0070] First, in step S120, the CPU 11 functions as the input acquisition unit 201 to acquire a medical image to be classified.

[0071] In step S122, the CPU 11 functions as the classification unit 202 and classifies each pixel of the medical image into either a normal class or an abnormal class using the trained model.

[0072] In step S124, the CPU 11, functioning as the classification unit 202, segments an abnormal region made up of pixels belonging to an abnormal class.

[0073] In step S126, the CPU 11 outputs the segmentation result by the classification unit 202 as the output unit 203. For example, for the medical image, an abnormal area consisting of pixels belonging to an abnormal class is displayed on the display unit 16, and the classification process is terminated. Then, for example, a doctor diagnoses the subject while looking at the segmentation result of the abnormal area for the medical image displayed on the display unit 16.

[0074] As described above, the learning device according to the present embodiment acquires a reference value for data belonging to a specific class among a plurality of classes, calculates the difference between the reference value and the learning data, and when the calculated difference is small, the learning data is given a lower importance than when the difference is large, and the more important the learning data is, the more importance is given to the learning data, and the model is learned. This makes it possible to learn a model while suppressing the variation in classification performance for ambiguous data.

[0075] For example, even if there is an ambiguous region in a medical image such as that shown in FIG. 12, the model can be trained while suppressing the variation in classification performance for the ambiguous region.

[0076] In addition, the classification device classifies input data into which of a plurality of classes the data belongs using a trained model trained by the learning device, thereby making it possible to suppress variation in classification performance for ambiguous data.

[0077] <Modification> The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit and scope of the present invention.

[0078] For example, although the example has been described in which the model is learned in the learning device 10, the model may be learned in the classification device 20. In this case, the classification device 20 may further include a learning data storage unit, an acquisition unit, a calculation unit, and a learning unit.

[0079] Although the above description is directed to a case where medical images are the target, the present invention is not limited to this example, and signals other than medical images may be classified as the target.

[0080] In addition, various processes executed by the CPU by reading the software (program) in each of the above embodiments may be executed by various processors other than the CPU. In this case, examples of the processor include a PLD (Programmable Logic Device) such as an FPGA (Field-Programmable Gate Array) whose circuit configuration can be changed after manufacture, and a dedicated electric circuit such as an ASIC (Application Specific Integrated Circuit) which is a processor having a circuit configuration designed exclusively for executing a specific process. Furthermore, the learning process and the classification process may be executed by one of these various processors, or may be executed by a combination of two or more processors of the same or different types (for example, a plurality of FPGAs, a combination of a CPU and an FPGA, etc.). Moreover, the hardware structure of these various processors is, more specifically, an electric circuit that combines circuit elements such as semiconductor elements.

[0081] In addition, in each of the above embodiments, the learning program and the classification program are described as being pre-stored (installed) in the storage 14, but the present invention is not limited to this. The programs may be provided in a form stored in a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The programs may also be downloaded from an external device via a network.

[0082] The embodiments disclosed herein are illustrative in all respects and should not be considered as limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the scope and meaning equivalent to the claims.

[0083] The following supplementary notes are further disclosed regarding the above embodiment.

[0084] (Appendix 1) A learning method for classifying input data into which of a plurality of classes the data belongs using a model, comprising the steps of: obtaining a reference value for data belonging to a specific class among the plurality of classes; calculating a difference between the reference value and learning data; When the calculated difference is small, the training data is assigned a lower importance than when the difference is large; A step of performing learning of the model by giving more importance to the learning data having a higher importance; A learning method in which a computer executes a process including the steps of:

[0085] (Appendix 2) 2. The learning method according to claim 1, wherein the data is sensor data, audio, an image, or a video.

[0086] (Appendix 3) 3. The learning method according to claim 1 or 2, wherein the plurality of classes includes a normal class and an abnormal class.

[0087] (Appendix 4) The learning method according to any one of Supplementary Note 1 to Supplementary Note 3, wherein in the step of acquiring the reference value, the reference value is acquired from data that does not satisfy a condition regarding a predetermined feature for data classified as belonging to a class other than the specific class.

[0088] (Appendix 5) 4. The learning method according to claim 1, wherein, in the step of acquiring the reference value, the reference value is acquired that is input from outside.

[0089] (Appendix 6) The learning method according to any one of Supplementary Note 1 to Supplementary Note 3, wherein in the step of acquiring the reference value, a representative value is calculated from a plurality of data belonging to the specific class that is prepared in advance, and acquired as the reference value.

[0090] (Appendix 7) the data is an image, The learning method according to claim 3, wherein in the step of acquiring the reference value, a maximum or minimum pixel value obtained from a region of the input data that is assigned in advance and that belongs to the abnormal class is acquired as a reference value for data that belongs to the normal class.

[0091] (Appendix 8) The learning method according to any one of Supplementary Note 1 to Supplementary Note 7, wherein in the learning step, if the calculated difference for the learning data is smaller than a threshold value, the importance is set to 0 and the model is trained.

[0092] (Appendix 9) The learning method according to any one of Appendix 1 to Appendix 7, wherein in the learning step, when the calculated difference is smaller than a threshold value for the learning data, the importance is reduced compared to data in which the calculated difference is larger than the threshold value, and the model is trained.

[0093] (Appendix 10) The learning method according to any one of Supplementary Note 1 to Supplementary Note 7, wherein in the learning step, a function that monotonically increases with respect to the magnitude of the difference is used for the learning data, and an output value of the function for the calculated difference is used as the importance to learn the model.

[0094] (Appendix 11) 8. The learning method according to any one of Supplementary Note 1 to Supplementary Note 7, wherein in the learning step, the model is trained so as to optimize a value of a loss function calculated based on the importance of the learning data.

[0095] (Appendix 12) The learning method according to any one of Supplementary Note 1 to Supplementary Note 7, wherein in the learning step, the model is trained using a gradient of backpropagation of an error calculated based on the importance of the learning data.

[0096] (Appendix 13) The learning method according to any one of appendices 1 to 12, wherein the data is a medical image obtained by capturing an image of a target part of a subject.

[0097] (Appendix 14) A step of classifying input data into which of the plurality of classes the data belongs, using a trained model trained by the training method according to any one of Supplementary Note 1 to Supplementary Note 13. A classification method in which a computer performs a process including:

[0098] (Appendix 15) the data is an image, 15. The classification method according to claim 14, wherein in the classifying step, for each pixel of the image, the trained model is used to classify which of the plurality of classes the pixel belongs to, thereby segmenting an area consisting of pixels belonging to at least one of the plurality of classes.

[0099] (Appendix 16) A learning device for classifying input data into which of a plurality of classes the data belongs using a model, comprising: an acquisition unit that acquires a reference value of data belonging to a specific class among the plurality of classes; A calculation unit that calculates a difference between the reference value and learning data; When the calculated difference is small, the training data is assigned a lower importance than when the difference is large; a learning unit that performs learning of the model by placing more importance on the learning data having a higher importance; A learning device comprising:

[0100] (Appendix 17) A classification unit that classifies input data into which of the plurality of classes the data belongs, using a trained model trained by the learning device according to claim 16. A classification device comprising:

[0101] (Appendix 18) A learning program for classifying input data into which of a plurality of classes the data belongs using a model, obtaining a reference value for data belonging to a specific class among the plurality of classes; Calculating the difference between the reference value and the learning data; When the calculated difference is small, the training data is assigned a lower importance than when the difference is large; The model is trained by giving more importance to the learning data having a higher importance. A learning program that causes a computer to carry out processing.

[0102] (Appendix 19) In the input data, a trained model trained by the learning program described in appendix 18 is used to classify which of the multiple classes the data belongs to. A classification program that causes a computer to carry out the processing. [Explanation of symbols]

[0103] 2. Classification System 10 Learning Device 11 CPU 14. Storage 20 Classifier 101 Learning data storage unit 102 Acquisition Department 103 Calculation section 104 Learning Department 201 Input acquisition unit 202 Classification Department 203 Output section

Claims

1. A learning method for classifying input data into which of a plurality of classes the data belongs using a model, comprising the steps of: obtaining a reference value for data belonging to a specific class among the plurality of classes; calculating a difference between the reference value and learning data; When the calculated difference is small, the training data is assigned a lower importance than when the difference is large; A step of performing learning of the model by giving more importance to the learning data having a higher importance; A learning method in which a computer executes a process including the steps of:

2. The learning method according to claim 1 , wherein the data is sensor data, audio, an image, or a video.

3. The method of claim 1 , wherein the plurality of classes includes a normal class and an abnormal class.

4. The learning method according to claim 1 , wherein in the step of acquiring the reference value, the reference value is acquired from data that does not satisfy a condition regarding a predetermined feature for data classified as belonging to a class other than the specific class.

5. 2. The learning method according to claim 1, wherein the step of acquiring the reference value acquires the reference value input from an external source.

6. 2. The learning method according to claim 1, wherein in the step of obtaining the reference value, a representative value is calculated from a plurality of data belonging to the specific class, which is prepared in advance, and obtained as the reference value.

7. the data is an image, 4. The learning method according to claim 3, wherein in the step of acquiring the reference value, a maximum or minimum pixel value obtained from a region of the input data that is assigned in advance and that belongs to the abnormal class is acquired as a reference value for data that belongs to the normal class.

8. 2. The learning method according to claim 1, wherein in the learning step, when the calculated difference is smaller than a threshold value for the learning data, the importance is set to 0, and the model is learned.

9. 2. The learning method according to claim 1, wherein in the learning step, when the calculated difference is smaller than a threshold value, the importance of the learning data is reduced compared to data in which the calculated difference is greater than the threshold value, and the model is learned.

10. 2. The learning method according to claim 1, wherein in the learning step, the model is learned for the learning data using a function that monotonically increases with respect to the magnitude of the difference, and the output value of the function for the calculated difference is set as the importance.

11. 2. The learning method according to claim 1, wherein in the learning step, the model is trained so as to optimize a value of a loss function calculated based on the importance of the learning data.

12. 2. The learning method according to claim 1, wherein in the learning step, the model is learned using a gradient of backpropagation of an error calculated based on the importance of the learning data.

13. The learning method according to claim 1 , wherein the data is a medical image obtained by capturing an image of a target part of a subject.

14. A step of classifying input data into which of the plurality of classes the data belongs, using a trained model trained by the training method according to any one of claims 1 to 13. A classification method in which a computer performs a process including:

15. the data is an image, 15. The classification method according to claim 14, wherein in the classifying step, a region consisting of pixels belonging to at least one of the plurality of classes is segmented by classifying each pixel of the image into which of the plurality of classes the pixel belongs using the trained model.

16. A learning device for classifying input data into which of a plurality of classes the data belongs using a model, comprising: an acquisition unit that acquires a reference value of data belonging to a specific class among the plurality of classes; A calculation unit that calculates a difference between the reference value and learning data; When the calculated difference is small, the training data is assigned a lower importance than when the difference is large; a learning unit that performs learning of the model by placing more importance on the learning data having a higher degree of importance; A learning device comprising:

17. A classification unit that classifies input data into which of the plurality of classes the data belongs, using a trained model trained by the learning device according to claim 16. A classification device comprising:

18. A learning program for classifying input data into which of a plurality of classes the data belongs using a model, obtaining a reference value for data belonging to a specific class among the plurality of classes; Calculating the difference between the reference value and the learning data; When the calculated difference is small, the training data is assigned a lower importance than when the difference is large; The model is trained by giving more importance to the learning data having a higher importance. A learning program that causes a computer to carry out processing.

19. The input data is classified as to which of the plurality of classes the data belongs to by using a trained model trained by the training program according to claim 18. A classification program that causes a computer to carry out the processing.

Images